Hacker News Reader

Comment by fyfy18 | Hacker News Reader

It's quite subjective, but my view is Matrix is what XMPP would like if it was written today.

The main advantages are it is using JSON instead of XML, and HTTP instead of custom protocols so it can work directly from browsers and with lightweight mobile clients. It also has a lot built into the core protocol, where as to get XMPP to do anything useful you will need to use a lot of extensions (and ensure the server and all clients support them).

The only con is it's still a relatively young project. I'm not sure I would want to rely on it as my main form of communication in a company just yet, but for personal usage it is fine.

View on HN

> It's quite subjective, but my view is Matrix is what XMPP would like if it was written today.

That's indeed a subjective view.

> The main advantages are it is using JSON instead of XML, and HTTP instead of custom protocols

The Matrix dev said a day ago that "It's NOTHING to do with HTTP+JSON": https://news.ycombinator.com/item?id=19217412

Also, what custom protocol are you talking about? XMPP works on top of TCP/TLS (or Websocket). Those are the IETF standards. In fact XMPP doesn't require an additional layer of HTTP at all.

> with lightweight mobile clients

Where are they? Anything even remotely comparable battery-wise with Conversations (an XMPP client)?

Not to mention that Matrix has a terrible server implementation which requires several GB of RAM just to join matrix.org room.

> It also has a lot built into the core protocol, where as to get XMPP to do anything useful you will need to use a lot of extensions (and ensure the server and all clients support them).

I fail to see how having a monolith core versus splitted specs has to do anything with code written to support that. Of course, initially every implementation will strive to implement full core (that was in XMPP in 2000s), but when the core evolves you have absolutely no guarantee that every developer will implement all new parts of the core. How is that addressed in Matrix?

> How is that addressed in Matrix?

Well that's quite easy, since hosting a server is a resource hog, most people are still on matrix.org, which gives it a special status in the ecosystem (even more special than jabber.org has been in the early 2000s fox XMPP). This allows them to "move fast and break stuff" on short notice (a few months), to bring the protocol forward and force people to upgrade. In the federated XMPP ecosystem, they essentially have to come to a community consensus between developers, servers operators and all other relevant actors to write a manifesto or a new spec that says "stop using $thing at $date".

In my opinion matrix, while being open-source, federated and easy to write clients for, is still a half-locked platform because essentially the matrix.org server is "too big to fail", and the matrix team is the only one who can e.g. offer consultancy over the protocol because nobody else can have a long-term view of its future evolutions (of course the matrix people are nice, and you can discuss with them, but that is still not a standards body).

> is still a half-locked platform... but that is still not a standards body

Exactly. That's why I always stress the importance of the IETF. The whole point of the IETF is that you cooperate with people, deal with opposite opinions, accept criticism, come to agreement, develop a compromise and produce a high quality absurdly documented standard. In Matrix bubble (and any others, see cryptocurrency or p2p community for example) they are the only one who makes the decision. They don't care on other opinions. If you disagree with them you're asked to stop "fighting", "focus on interaction", "understand their goals" and other meaningless stuff like "let the best standard wins" (when in reality the whole point of the standard is to cooperate). Formally speaking they try to substitute a "standard document" with an "open document" and people kinda swallow this because they don't see the difference.

this is categorically not true.

> In Matrix bubble (and any others, see cryptocurrency or p2p community for example) they are the only one who makes the decision. They don't care on other opinions.

You can see our governance model at https://github.com/matrix-org/matrix-doc/blob/matthew/msc177... and see that we do everything we can to ensure that the wider community can contribute to the spec and steer it in the right direction.

And fwiw, we'd be perfectly happy to contribute the spec to IETF or W3C or whoever once it's relatively stable, just like XMPP did.

Rather than burning all this energy spreading FUD about Matrix, perhaps it might be better spent improving XMPP? :/

> Rather than burning all this energy spreading FUD about Matrix, perhaps it might be better spent improving XMPP? :/

Yeah, didn't I say above about such kind of arguments?

> You can see our governance model at https://github.com/matrix-org/matrix-doc/blob/matthew/msc177.... and see that we do everything we can to ensure that the wider community can contribute to the spec and steer it in the right direction... And fwiw, we'd be perfectly happy to contribute the spec to IETF or W3C or whoever once it's relatively stable, just like XMPP did.

I'm not a lawyer to read such documents. Just "contribute" your core to the IETF, prove me wrong.

That's when committees and standards bodies work well.

When they don't, you get OAUTH2.

https://hueniverse.com/oauth-2-0-and-the-road-to-hell-8eec45...

I might be mistaken, but from what I read the main addition is that all peers replicate the full state.

If you have a dominant xmpp server (like gmail was) that removes xmpp support, its users have the start using other clients to connect to it or start a new social structure from scratch. In essence this makes the xmpp server a dominant player with leverage.

If gmail had used matrix when it pulled the plug all interactions (who knows who, how to route traffic, chat history) would have been replicated at the peers so it would just have removed google’s servers from the equation. The people who had been using it would continue uninterrupted.

You're both right and wrong.

In both cases, if users have an account on gmail and are using it, and gmail pulls the plug, then they can't use it anymore and can't migrate. The history is still there in the case of matrix, but they can't reuse the same account.

However if another user is homed on another server, they would still keep the knowledge and history on their side: gmail did pull the plug on xmpp, yet people are still using it. Their server still know who their contacts are, what room they're usually in, etc...

Matrix does have the advantage that the chat history is fully replicated on all homeservers so it makes history retrieval much saner, but that's it. Matrix doesn't route traffic (messages go from sender's homeserver to recipients' homeservers, that's it) and doesn't share client's information with the whole world.

Too bad. Well consider this a feature request then ;)

> The people who had been using it would continue uninterrupted.

I'm not sure about that. If the main matrix server disappeared today that'd be problematic. Yes, you'd have history locally but the same can be said for any native client that stores history locally.

As far as I know there is no way to migrate matrix history now. Is there? (genuinely curious)

for rooms which span servers, the history is already migrated by nature of the protocol being decentralised.

for everything else we’re working on decentralised accounts via MSC1228.

there are also migration scripts and tools to move you between accounts on different servers today (like email).

> I'm not sure about that. If the main matrix server disappeared today that'd be problematic

This ^^ Currently a large amount of users (I'm told around a half, but I'm not sure) have matrix.org as their home server. So Matrix is effectively building a power-law network with all its drawbacks: when the "power node" of the network goes down the whole network falls apart.

To be fair, this is somewhat inevitable in the initial stages, before a reasonable spec is released. Many people, like me, are just "sitting on the fence" and test-driving Matrix via their matrix.org server, until there are reasonable assurances that deploying our own Matrix server will come without too much hassle.

After Synapse 1.0 is released (sometimes next month, IIRC), I'm definitely deploying my own homeserver, and federating it with the rest of the world.

FWIW, I've been running Synapse since 0.20 or so (for at least 2 years now), and there has never been a breaking change in the protocol as far as I'm aware. I had to read release notes at some points to get the upgrade to work, but it was mostly Python-library-related stuff IIRC.

Well they are making one big instance nothing else. Also unfortunately it means that you can't possibly federate with matrix.org. I've been running small otherwise perfectly functioning instance for few years for my circle/work as slack alternative on small vps. One time i did the amazing mistake of turning on federation with matrix.org. It immediately killed my servers performance especially when i joined some popular channel. The channels probably have more messages than my whole server combined. And it works the other way too - matrix.org is like sponge sucking other instances.

I guess this is the real problem for me - i would love to federate but i can't possibly police or enforce what servers/channels are small enough that my users can join them.

This isn’t anything to do with the size of the matrix.org server; just that synapse needs to be more resource efficient and we need to limit the sizes of rooms small servers try to join. we are doing both.

I can't see any advantage in using JSON and HTTP for a protocol. JSON shines for ad-hoc, untyped, structured message payloads rather than the strongly-typed semistructured text payloads possible with XML/SGML, and HTTP wasn't designed for federated protocols. It seems more like a fashion thing to me.

I kind of agree with you, but I'll add that XMPP usage of XML is weird enough that it makes it hard it use a normal XML parser / producer. Honestly I really wanted to like XMPP. I read all the specs at the time, even started to write some code, but all I can say is that it really sucks. I haven't looked at matrix yet, I hope it's better

Indeed! I was in the same boat - also started writing a client, but wasted too much time on the stupid XML stream idea.

Basically, chatting is message oriented, but the XMPP inventors thought it would be neat to start the connection with a tag like <beginconversation> which would then be dangling for the whole connection, preventing the use of a simple DOM XML parser (remember this is many years ago) for each message.

Any XML parser that I know can parse XML in a stream, that's all you need to handle an incoming XMPP stream.

Just an example, here is the XML parser that I wrote in PHP https://github.com/movim/movim/blob/master/lib/moxl/src/Moxl... (127 lines), it basically handle the stream as an input and fire SimpleXML (can also be DOM XML) on the other hand.

This can handle thousands of XMPP messages (called stanza) per second :)

Can it handle plaintext XML suddenly becoming TLS encrypted? Because that's also what happens on any XMPP connection. If it can, then great but I guess it's more an exception than the rule

fwiw, in Matrix, to send a message, it's:

  curl 'https://matrix.org/_matrix/client/r0/rooms/!RXNcERlITUVszBNIPW%3Amatrix.org/send/m.room.message/123?access_token='$TOKEN -X PUT --data '{"msgtype":"m.text","body":"hello world"}'

and to receive messages it's:

  curl 'https://matrix.org/_matrix/client/r0/sync?access_token='$TOKEN

Hopefully this ends up being quite easy to work with :)

and for XMPP

    <message to="user@server.com"><body>Hello world</body></message>

No need for a token, the socket you're in is already authenticated :)

Also flexibility can be good but XMPP is way too modular.

You don't have to implement all the modules :p There is some basic ones that you need (see XEP-0412: XMPP Compliance Suites 2019 - https://xmpp.org/extensions/xep-0412.html) and you have already an up to date client and server. The rest if for specific use cases.

Part of their rationale is explained in their FAQ: https://matrix.org/docs/guides/faq#why-http%3F-doesn't-http-...

I dislike JSON, but it's a step in the right direction from XML. XML is an inelegant, verbose, over-specified mess, a relic of 90s thinking. JSON is definitely an improvement.

HTPP's fine, I guess. It's certainly available everywhere, and saves one from writing a bunch of tiresome parsing code. But that's really an indictment of the current state of our technology. Popular languages should make it easy to read structured data out of a socket (with some sort of max length to prevent DDoSes) & write structured data to a socket. Since they don't, HTTP is an okay compromise. It's a bit heavyweight, but it gets the job done.

Reply on news.ycombinator.com